@agentpay-run/mcp-server

Official MCP server for AgentPay.

Lets agents running through any MCP-compatible client (Claude Desktop, Cursor, Continue, your own MCP host) charge vendors, check budgets, and inspect tier quota through your AgentPay wallet — with all of AgentPay's policy gates, anomaly detection, and audit logging in front.

Why

The state of the art for giving an agent a budget is "share your Stripe key and pray." AgentPay puts a real spend-control layer between the agent and the payment rail. This MCP server exposes that layer as standard MCP tools so any agent can use it without a vendor-specific integration.

Tools exposed

| Tool | What it does | |--------------|-----------------------------------------------------------------------------------------------| | charge | Charge a vendor. Returns approved/denied + which policy matched + remaining budget. | | get_wallet | Read-only wallet snapshot: budget, spent, per-tx cap, vendor allowlist, active state. | | get_quota | Read-only tier + remaining quota: wallets, today's charges, audit retention. |

Install in Claude Desktop

1. Get a wallet API key at agentpay.run/dashboard (use a ap_test_... key while you wire things up).
2. Open claude_desktop_config.json (Claude Desktop → Settings → Developer → Edit Config).
3. Add:

{
  "mcpServers": {
    "agentpay": {
      "command": "npx",
      "args": ["-y", "@agentpay-run/mcp-server"],
      "env": {
        "AGENTPAY_API_KEY": "ap_test_..."
      }
    }
  }
}

4. Restart Claude Desktop. You'll see the agentpay server in the MCP indicator (bottom right of the chat).

Install in Cursor / other MCP hosts

Anywhere that accepts an MCP server config, point it at npx -y @agentpay-run/mcp-server with the same AGENTPAY_API_KEY env var.

Usage

Once installed, just talk to your agent like you would otherwise:

"Charge openai.com $12 against my dev wallet, then tell me what's left."

The agent picks charge, reads back the response, and decides what to do based on the policy verdict.

Configuration

| Env var | Required | Default | Notes | |----------------------|----------|------------------------|--------------------------------------------------------------------------------------| | AGENTPAY_API_KEY | yes | — | Wallet API key. ap_test_... for development, ap_live_... for production wallets. | | AGENTPAY_BASE_URL | no | https://agentpay.run | Override for self-hosted or local dev (e.g. http://localhost:3000). |

Local development

git clone https://github.com/sdeanesh-png/agentpay
cd agentpay/sdk/mcp
npm install
npm run build
AGENTPAY_API_KEY=ap_test_... node dist/index.js

The server speaks MCP over stdio. To smoke-test it manually, pipe a JSON-RPC tools/list request:

echo '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'   | AGENTPAY_API_KEY=ap_test_dummy node dist/index.js

Security

• The wallet API key is the only secret. It's wallet-scoped — even if compromised, the blast radius is bounded by the policy you set on that wallet (per-tx cap, monthly budget, vendor allowlist).
• The server is stateless. Every call re-queries AgentPay; nothing is cached locally.
• For production, use ap_live_... keys with strict per-tx caps and a vendor allowlist. Treat them like API tokens: rotate via the dashboard if exposed.

License

MIT.